Fake News Detection Using Transformer-Based Models with Explainable Artificial Intelligence

The rapid proliferation of misinformation on online platforms poses a serious threat to public trust and information integrity. Although transformer-based models achieve state-of-the-art performance in natural language processing, their black-box nature limits deployment in high-stakes domains such as fake news detection. This paper proposes an interpretable end-to-end fake news detection framework that integrates a fine-tuned Bidirectional Encoder Representations from Transformers (BERT) model with Explainable Artificial Intelligence (XAI) techniques. The proposed pipeline includes data preprocessing, transformer-based classification, and posthoc interpretation using SHAP and LIME. Experiments conducted on the ISOT and LIAR benchmark datasets demonstrate that the proposed model achieves accuracy scores of 98.2% and 92.6%, respectively, outperforming several traditional machine learning and deep learning baselines. Furthermore, the explainability analysis reveals meaningful linguistic patterns influencing model decisions, thereby enhancing transparency and user trust. The results indicate that the proposed framework effectively balances predictive performance and interpretability, making it suitable for real-world misinformation detection systems.